Oswlek

Every year for Pats playoff games, I look at how NE and their opponent scored on and defensed their opposition in relation to how those teams did. (Simply put, scoring 21 points on Baltimore is more impressive than scoring 28 on Detroit - just doing this same thing for all teams faced by both teams). Since I have started this in 2003, it has been right every single time, regretably even last year. I feel that points scored is a good reference point because it gives credit to teams that can score on defense and special teams.

In 2003 and 2004 I used totals for the entire season, but like last year, I am not going to do that this year, at least for the Jet game. In this matchup, I think it makes more sense to count from the previous matchup on. For you lurking Jet fans, I am doing this partially because NE has improved, but primarily because NY has been a totally different team since their bye week. Doing it this way actually works in NY's favor.

Since NY's bye week (including the NE game, which obviously works even more in NY's favor) the Jets the following details:

PF - 19.5
PA - 12.8

The average of their opponents' season averages are:

PA - 19.4
PF - 18.7

The Jets have scored .1 more than their opposition has allowed, on average (19.5-19.4) and allowed 5.9 less for a combined score of +6.

Now that I think my methodology is clear, NE has the following breakdown from the Jet game on:

PF - 23.5 vs 20.6 allowed by the oppositing ~ +2.9
PA - 15.4 vs. 20.1 scored by opposition ~ +4.7
Total score ~ 7.6

The two are closer than I thought when I started on this venture, but NE is clearly ahead and that is all that has mattered in the past. This is good news. It is interesting that it is the Jets, not the Pats that have held the opposition further below their averages during this time.

Furthermore, both teams had pretty ugly losses 4 weeks ago and closed with three straight wins. Over the two team's respective streaks:

NY
PF - 20.7 vs. 19.6 ~ +1.1
PA - 8.7 (!) vs. 14.8 ~ +6.1
Total score ~ 7.2

NE
PF - 34.7 (!) vs. 21.7 ~ +13
PA - 17 vs. 20.1 ~ +3.1
Total score ~ 16.1

As impressive as the Jets' total points allowed over their streak is, it is only marginally better than how they had been doing in the games prior. They just played three dreadful offenses. It still was better than NE's, though.

What we have to hope for is that Pony's "turncoat" factor doesn't raise its ugly head, because NE has just been killing defenses lately.

Another thing to consider is the fact that NE's final two opponents had legitimate playoff hopes while NY faced two teams long out of it and on other running on fumes.

I put more stock in the numbers over the longer term, but everything seems to indicate to me the following odds:

20% Jets win a close game
45% NE wins a close game
20% NE wins a moderately close game, but the game was rarely in doubt
15% NE wins in a blowout


Edited to add that I ignored home/away factors because both played two away games and the weakest opponent for both was at home. Also, I used full season totals in calcuating my averages, not the numbers for those teams over the same time. If anything, that helps the Jets, but only marginally because both played teams (Miami & TN) that were better later in the year than before.

Oswlek

I guess this is a little dry for some folks.

Bump to see if anyone else is interested.

PonyExpress

Fascinating. How has this method worked in the past? It would be interesting to go back and check. Do you think a pure points number is the best way to analyze, or maybe a percentage under a mean? For instance, the Pats D kept opponents to -25% under scoring average, the Jets D -33%, etc.

Oswlek

Since 2003 it has correctly predicted every game that I have done, although I will admit that I don't think NE would have been forcasted to win any of the 2001 games, save for maybe the Oakland game.

It is not that much help in predicting spreads, however, just victors. The closest of any of the prior matchups was the Philly SB. The biggest disparity was the Carolina SB.

I have not ever looked into breaking it into %'s partially because I don't have the time and partially because there has been no need. I can see how it might be useful, though.

psychoPat

PatsFanSince74

Thanks for a lot of work and a lot of thinking.

The problem I have is that freezing this kind of analysis (no matter how excellent it is, and your's is certainly excellent) and applying it to one, specific game gets very tricky. And, the fewer data points you have (in this case a handful of games by each team) the more volatile any prediction about a single game using the data becomes.

IMHO, a flaw in Carroll's brilliant and groundbreaking Work ("Hidden Game of Football," with which I am sure you are familiar) is that it falls victim to what statisticians like to call the "Flaw of Averages" and doesn't allow sufficiently for the distribution of a series of possible outcomes at any point on the field or, by extrapolation, in a game. (This is something that I think the DVOA analysis is trying to correct, but I have to admit I haven't studied it sufficiently yet to be sure.)

So, as with any analysis of this nature, I think it's important to acknowledge that the Standard Deviation of the various outcomes around what we may conclude is a likely outcome is very high. I'm comfortable with what you're saying up to the point of concluding that there's a 65% chance that this will be a very close game and that, within that probability, it's more likely that the Patriots would win than the Jets. I part with you in trying to use the data to conclude that the odds of the latter among close outcomes are are better than 2--1, as you suggest, and that there is no possibility of a Blowout in the other direction, which you probably omit for the same reason that I would--it's unthinkable.

But since you've had the guts to say what you think, here's what I think:
Patriots win in a Blowout, defined as more than two possessions: 10%
Patriots win a moderately close game, defined as two possessions or less but in which the outcome was never really in doubt:15%
Patriots win a close game, defined as a single possession game in doubt until the final possession: 35%
Jets win a close game, defined as a single possession game in doubt until the final possession: 25%
Jets win a moderately close game, defined as two possessions or less but in which the outcome was never really in doubt:10%
Jets win in a Blowout defined as more than two possessions: 5%

So, I'm saying there's a 60% chance that the Pats win, that there's a 60% chance that the game will come down to the final possession with the Pats having the edge, that there's a 25% chance that the Pats will win a game that's never in doubt, and that there's a 15% chance the Jets will win a game that's never in doubt.



And, no matter how I analyze it, I'm still going to feel like throwing up at the opening kickoff and will be a frenzied idiot until the game is resolved.

Thanks again for posting this.

__________________
It is what it is. It wasn't what it wasn't.

Brownfan80

Os, interesting methodology. And even more interesting that it's been correct so often!! Hopefully the trend continues (at least until the next time it predicts a Pats loss, then the method better be wrong).

The Dude

Great stats. Mirror the accuscore 80 - 20% prediction for the game.

I am sure glad I found this site and can get away from the idiots over at the boston globe

Oswlek

Alright. I just finished doing a little backtesting. I would have liked to have done more, but it will have to do for now. I looked at last year's AFC playoff field. Since I already knew that NE rated higher than Jax and lower than Denver, I didn't bother to calculate NE's or the Jags' scores. Shoot me. Here is the damage:

Pitt:
PF - 24.3 vs 20.1 ~ +4.2
PA - 16.2 vs. 20 ~ +3.8
Total ~ 8

Pitt with Roethlisburger
PF - 25.4 vs. 20.5 ~ +4.9
PA - 15.7 vs. 22 ~ +6.3
Total ~ 11.2

Cincy
PF - 26.3 vs. 19.6 ~ +6.7
PA - 21.9 vs. 19.2 ~ +2.7
Total ~ 9.4

Indy
PF - 27.4 vs. 21.9 ~ +5.5
PA - 15.4 vs. 20.8 ~ +5.4
Total ~ 10.8

Indy ignoring final two games
PF - 29.2 vs. 22.1 ~ +7.1
PA - 14.7 vs. 20.3 ~ +5.6
Total ~ 12.7

Denver
PF - 24.7 vs. 19.4 ~ +5.3
PA - 16.1 vs. 21.4 ~ +5.3
Total ~ 10.6

Conclusions/Observations
* I think it is fair to say that I would have used the "with Ben" figure for Pitt, which makes their victories over Cincy and Denver no flukes. Without even doing this breakdown, I did wager on Pitt in that game. Of course I also put money on NE against Denver

* I would have probably used the Indy one that excluded the final two games. However, with Dungy's son and Indy's playoff history it is not surprising that Pitt was able to win that game. (In hindsight anyway. I was surprised at the time.) Had I done this last year - I only did it for NE games - I would have realized that these two teams match up more closely than they initially appeared.

* Considering NE was 6-0 in 2003 and 2004 using this measure (and I knew that they would beat Pitt big because they had a very large advantage - yes I know I just said before that spread doesn't correlate as well with this, but I hate them dammit!), in the games that I have tracked so far the team with the best measure is 10-1. And the one outlier is Indy.


Edited to add that even assuming Indy's numbers should have excluded the last two games, the gap of 1.5 between Pitt and Indy's numbers is actually smaller than NE and NY's (1.6). Not only that, but Indy did not have a sizable end of the year trend up that theoretically increased that gap. Based on this, it would actually be a bigger upset of NY wins than it was when Pitt beat Indy last year.

PatsFanSince74

Yeah. The application of statistics and analytics to the NFL isn't anywhere near the sophistication of MLB yet, but it's getting better all the time. The darn problem is that in any given NFL season each team can only contribute around 60 plays per game 16 times to the database and none of it gets truly meaningful until the end (so 16 games times 11 Players times 60 Plays times 32 teams equals 337,920 datapoints). Contrast that with the MLB where each team contributes a minimum of 27 datapoints each game over 162 games (27 minimum at bats per game times 162 games times 9 players times 30 teams equals 1,180,980 minimum data points).

__________________
It is what it is. It wasn't what it wasn't.